Defining molecular vulnerabilities in childhood leukaemia through biological network analysis

Cellular phenotype is largely governed by the regulation of gene expression that dictates critical biological processes such as cell differentiation, adaptation to stimuli, and metabolism. Gene regulation is typically altered in cancer, leading to dysregulation of networks involving multiple genes, transcription factors and genomic regulatory elements. Studying these gene regulatory networks (GRNs) can help us untangle the regulatory mechanisms that underpin disease biology and potentially identify new therapeutic targets. The aim of the research reported in this thesis was to analyse gene regulation in cancer in order to identify new vulnerabilities and understand mechanisms of gene regulation and resistance to current therapies. Firstly, we took a systematic approach to evaluate the ability of GRNs to predict gene essentiality in cancer cell lines. We employed computational methods to infer the activity of regulatory genes in cell lines from ten different cancer types. We then tested the ability of GRN-inferred activity to predict the sensitivity of different cancer cell lines to gene inhibition, using genome-wide CRISPR screens from The Cancer Dependency Map. We found that while GRNs display some cancer specificity, GRN-inferred activity does not perform any better than gene expression at finding essential genes in any tumour type. Treating sensitivity to inhibition as a binary variable or assessing the ability of GRN-inferred activity to predict gene essentiality led to similar results, with gene expression performing better than inferred activity. Finally, stratifying GRNs by their size or number of unique targets did not improve predictions for GRN-inferred activity. Our results were concordant across multiple GRN sources and activity estimation methods. Secondly, we took a focused approach to study genetic and epigenetic regulation in the blood cancer acute myeloid leukaemia (AML). To replicate PRC2 loss-of-function, which is present in 15% of paediatric AML and which is linked to chemoresistance, we used CRISPR/Cas9 editing to create isogenic cell line models of heterozygous EZH2 loss. We found that EZH2+/- AML cells had altered gene expression, decreased genome-wide repressive chromatin marks, and notably had markedly increased chromatin accessibility. This altered regulatory landscape resulting from the depletion of a genome-wide epigenetic transcriptional repressor led to partial activation of alternative lineage transcriptional programmes, including overexpression of the fetal haematopoiesis gene LIN28B. The activation of a LIN28B-driven programme included increased CDK6 expression that correlated with decreased sensitivity to CDK6 inhibition in EZH2+/- cells. Interestingly, the 3D genome architecture was largely maintained upon EZH2 depletion, with preferential retention and even gain of H3K27me3 at regions with high 3D contact frequency. Overall, our work provides insights into approaches to studying gene regulation and applications of computational methods to understand this field. In addition, we provide a detailed characterisation of the complexities of genomic regulation in PRC2-depleted AML that has implications for understanding the aggressive disease biology in these cases.